competitor-market-watch

name	competitor-market-watch
description	Generate the user's recurring competitor and market-watch brief from public sources. Track named competitors and market signals - pricing, launches, changelogs, positioning, hiring, reviews, news - detect what changed since the last run, filter against what could move the user's roadmap, pricing, sales talk track, or week plan, and deliver a short brief tagged watch / ignore / act. Read-only, every claim source-cited. Improves from feedback.
version	0.1.0
metadata	{"hermes":{"tags":["competitive-intel","market","briefing"],"category":"reports"}}

Competitor & Market Watch

TEMPLATE - NOT YET BOOTSTRAPPED. This skill ships as a workshop template. While this banner exists, do not generate a brief. Run the Bootstrap section below instead: interview the user, then edit this file so the skill belongs to them. Bootstrap is finished when this banner and the Bootstrap section have been deleted.

Watch a handful of competitors and market signals so the user doesn't have to open five tabs every Monday:

1. Read the same public sources every run - sites, pricing, changelogs, blogs, reviews, news.
2. Compare against last run (see Memory) to find what actually changed.
3. Keep only changes that could move the user's roadmap, pricing, sales talk track, or plan.
4. Deliver a short brief - each item tagged watch / ignore / act, with a source link.

The value is not "here's what competitors are doing." It's "here's what changed that matters to you, and whether to act on it."

Bootstrap (agent: run once, then delete this section)

Interview the user - one question at a time, rough answers welcome. Don't over-collect; the feedback loop sharpens everything later. Four questions.

1. Who are you up against? 2-5 competitors or alternatives, plus one line on what the user sells and to whom. That last part is what makes the brief relevant. Or, just give me your website and I will come back with a list of your competitor for you to approve.
2. Which public sources per competitor? Pricing/product pages, changelog or release notes, blog, app-store or review pages, job posts, news. One to three each to start - URLs are best. If they don't know, suggest starters and let them pick.
3. What counts as a change worth surfacing, and what's noise? e.g. surface pricing, packaging, launches, positioning shifts, hiring for a capability; ignore press fluff, recycled announcements, funding gossip, rumor with no source.
4. What would make you act, and how often should this run? The decision filter (roadmap, pricing, sales talk track, week plan) and the cadence (weekly is usually plenty).

Then edit this skill file (skill_manage edit, or edit the file directly):

• Replace Who I watch with the user's real competitors, sources, and their one-line "what I sell / to whom". Record a concrete URL for every source so each run fetches directly instead of searching.
• Fold the user's answers into What counts as a change (and its Drop list) and Decision filter.
• Rewrite the frontmatter description so it names the user's actual watch.
• Leave the workflow, memory, brief format, output, and feedback sections alone unless the user asked for something specific.
• Delete this Bootstrap section and the template banner at the top.

Finish with two steps:

1. Offer one short message the user can send right now to generate their first brief. Remind them the first run is a baseline - nothing to compare against yet, so it maps the current state; change-detection kicks in on run two.
2. Check whether a Hermes gateway is configured (Telegram, Discord, email). If one is, set up a Hermes cron on the cadence the user chose, delivering the brief through that gateway. If not, tell the user they can connect a gateway later and schedule it then.

Who I watch

What I sell / to whom: a self-hosted team wiki, sold to engineering teams at 20-200 person startups who want to own their docs instead of renting a SaaS.

1. Northwind Docs - hosted wiki, our closest head-to-head.
   • Pricing: https://northwind.example.com/pricing
   • Changelog: https://northwind.example.com/changelog
   • Blog: https://northwind.example.com/blog
2. Contoso Notes - bigger, moving down-market toward our buyer.
   • Pricing: https://contoso.example.com/pricing
   • Release notes: https://contoso.example.com/releases
3. The category in general - "self-hosted wiki" / "Notion alternative" launches and comparisons in news, Hacker News, and relevant subreddits.

What counts as a change

Surface an item only if it is a change since last run and it could touch the user's world:

• pricing or packaging moves (new tier, price change, feature moved between tiers);
• launches, major features, or deprecations in a competitor's changelog;
• positioning or messaging shifts (homepage, category they claim, who they target);
• hiring signals for a capability the user competes on;
• review-volume or sentiment shifts on app-store / review pages;
• market news - funding that changes a rival's trajectory, entrants, acquisitions.

Drop:

• press fluff, recycled or re-announced news, rumor with no source;
• generic AI hype and vendor marketing;
• anything unchanged since the last brief (see Memory) - don't re-report the baseline;
• items that touch none of the user's decision filters below.

Decision filter

Only surface a change if it could plausibly move one of these:

• Roadmap - a feature we'd now consider building, dropping, or reprioritizing.
• Pricing / packaging - pressure to change our tiers or price.
• Sales talk track - something a prospect will bring up, that we need an answer for.
• This week's plan - anything urgent enough to act on now.

If a change is real but touches none of these, tag it watch and keep it to one line - or drop it.

Run workflow

1. Get today's date with a tool.
2. Read Memory (state.md next to this skill) - what each source looked like last run. On the very first run there is no file yet; that's fine, this run is the baseline.
3. Fetch the sources in Who I watch. Public pages only - no logins, no scraping behind auth. Prefer official/direct pages (pricing, changelog, blog) over broad search; search only to fill gaps or for the "category in general" signal.
4. For each source, diff against Memory: is this new, a change to something known, or unchanged? Unchanged items are dropped from the brief (they stay in Memory).
5. Keep changes that survive What counts as a change and the Decision filter. Tag each watch / ignore / act.
6. Write the brief in the format below. Every factual claim carries a source link. If a competitor was quiet, say so in one line - never pad.
7. Deliver per Output.
8. Update Memory: record the current state of every source you checked so next run can diff against it.

Memory (state across runs)

Change-detection needs a record of what things looked like before. Keep one plain-Markdown file, state.md, next to this skill:

• One section per competitor / source.
• Under each, the facts you'll want to diff next week: current pricing/tier structure, the newest changelog entry (date + title), current homepage tagline, latest review count - whatever the user cares about.
• A last checked: YYYY-MM-DD line per section.

Each run: read state.md first, diff against it to find changes, then overwrite each section with the current state at the end of the run. Keep it terse - it's a snapshot for diffing, not a log. If it doesn't exist yet (e.g. a curl install of only SKILL.md), create it on the first run.

Brief format

Example shape - your real competitors and changes replace the placeholders:

# Competitor & Market Watch - YYYY-MM-DD

## What changed
1. **[act] Competitor A added a $99 team tier** - undercuts our Team plan; sales will hear
   about it. · [pricing](link) · *Suggested: draft a talking point for the Team comparison.*
2. **[watch] Competitor B shipped SSO** - not our buyer's blocker yet, but trending that way.
   · [release](link)

## Quiet
- One line per source that had no material change since last run.

## Tune-up
- One suggestion: a source, competitor, or filter to add or drop.

Aim for 3-7 items, each a real change - short enough to read with coffee. The first (baseline) run maps the whole landscape, so it runs a little longer; after that, unchanged sources drop to Quiet and most weeks are short. Honest when it was a quiet week - a one-line "nothing material moved" is a valid brief.

Output

• Default: Markdown in chat. No files. When a gateway is connected (Telegram, Discord, Slack, email), "chat" means that gateway. Only write files or send anything else if the user asked.

Feedback loop

After each brief, ask:

• What was useful? What was noise? What was missing?
• Any competitor or source to add or drop?
• Did the watch / ignore / act tags match your judgment?

Then edit this skill with the answers: competitors and sources → Who I watch; what to surface or ignore → What counts as a change; what makes you act → Decision filter; format changes → Brief format. This is the whole self-improvement loop: feedback in, skill edits out, next brief sharper.

Later

If no gateway was connected at bootstrap, schedule the brief with Hermes cron once one is; deliver it over Telegram / Discord / Slack / email; or split into per-competitor watches feeding one combined brief.